Katanin p60-like 1 sculpts the cytoskeleton in mechanosensory cilia

Mechanotransduction is a cellular process that converts external mechanical stimuli into intracellular signals (Gillespie and Walker, 2001; Lumpkin et al., 2010). It is the key mechanism underlying the perception of touch, sound, and acceleration. In mechanotransduction, external forces impinging on sensory cells are thought to be directly coupled to force-sensitive channels (Corey and Hudspeth, 1983; Gillespie and Walker, 2001), for which a mechanical pathway is thought to be formed by cellular structures, including the extracellular matrix, membrane, and cytoskeleton (Gillespie and Walker, 2001). Among these components, cytoskeletons are often found at the site of mechanotransduction; for example, F-actin bundles in the stereocilia of vertebrate hair cells (Gillespie and Müller, 2009), microtubules in the nonciliated touch receptors of Caenorhabditis elegans (Cueva et al., 2007), and in fly ciliated mechanoreceptors (Bechstedt et al., 2010; Keil, 1997). Because these cytoskeletons play both structural and sensory roles in mechanosensation (Cueva et al., 2007; Liang et al., 2014), how they are organized and formed is important in understanding the mechanism of cell mechanotransduction.

External mechanosensory (ES) organs, including campaniform and bristle receptors, belong to type I sensory receptors of Drosophila melanogaster (Keil, 1997). In ES organs, the dendrite of the sensory neuron is separated into the outer segment located at the distal tip and the inner segment at the proximal part by a short “9+0” cilium (Keil, 1997; Liang et al., 2013). The outer segment is a cilium-based entity that contains a mechanoreceptive organelle (MO) and a tubular body (TB). There is a large quantity of microtubules inside the outer segment (Bechstedt et al., 2010; Keil, 1997). These microtubules show a relatively high level of acetylation (Liang et al., 2011) and form a specialized cytoskeleton (Bechstedt et al., 2010; Keil, 1997; Liang et al., 2014), which structurally supports the sensory cilium and functionally contributes to mechanotransduction (Liang et al., 2014). However, the structural and molecular basis of this cytoskeleton remains unknown. First, the ultrastructural organization and stability of these microtubules have not been characterized. Second, how the cytoskeleton is formed remains elusive. A previous study on dcx-emap suggests the presence of specific regulatory proteins in the outer segment (Bechstedt et al., 2010). However, their identity and working mechanism remain unknown.

Campaniform mechanoreceptors are located at joints (e.g., wing base; Keil, 1997; Liang et al., 2017), and they respond to various types of cuticular strains caused by internal (e.g., muscular activity) or external (e.g., mechanical compression) mechanical stimuli to provide sensory feedbacks (Pringle et al., 1948). The characteristic dome-like external structures and their localizations near the cuticular surface facilitate the study of these receptors using electron and light microscopy (Liang et al., 2013; Sun et al., 2019a). Furthermore, the high similarity among different ES organs makes it possible to use campaniform mechanoreceptors as a model to study the sensory neuron in fly ES organs.

In the present study, we resolved the 3D structure of the cytoskeleton in the outer segment of campaniform mechanoreceptors using electron tomography (ET) and showed that this cytoskeleton is formed of mostly unidirectional and stable microtubules using light microscopy. Furthermore, we identified that katanin p60-like 1 (kat-60L1), a neuronal type of microtubule-severing enzyme, contributes to mechanosensitivity and sculpts the cytoskeleton in the mechanosensory cilia. Structural analyses on mutants suggest that kat-60L1 amplifies the mass of microtubules to form dense microtubule arrays and generates short microtubules to build the nanoscopic cytoskeleton in the MO. Further analyses reveal the functional roles of Patronin and other potential regulatory factors in promoting microtubule assembly and stability in the outer segment. In all, our results characterize the specialized cytoskeleton in fly mechanosensory ES organs in detail and provide mechanistic insights into how the cytoskeleton is formed.

We set out to reconstruct the outer segment of campaniform mechanoreceptors using a recently developed protocol based on ET (Sun et al., 2019a). Two types of campaniform mechanoreceptors were analyzed: the arrayed receptors in the pedicel of halteres (Fig. 1, A–I) and the isolated receptors in legs (Fig. 2, A–F; and Fig. S1). Despite the differences in detailed geometry, the outer segment in both types of cells was separated from the inner segment by a short cilium, and it contained an MO and a TB (Fig. 1, B and I; and Fig. 2, B and D). To better understand the cytoskeletal structure, we traced the microtubules in the outer segment and built the 3D model (Fig. 1 I, Fig. 2 D, and Video 1).

In haltere pedicellar receptors, the TB had a nearly spherical morphology (x axis, 1.4 ± 0.7 µm; y axis, 1.4 ± 0.7 µm; z axis, 1.2 ± 0.6 µm; n = 3 receptors). Nine microtubule doublets were curved to fit the dilated shape of the TB and embraced an array of interconnected microtubules inside the TB (Fig. 1 I). By inspecting the positions of microtubule ends that appeared within the volume of tomographic reconstruction (Fig. S2), we found that microtubules in the TB were short (L_{TB_MT} = 296 ± 191 nm, n = 2,524 microtubules from four cells). They were mostly aligned to the proximal–distal axis, but some transversely oriented ones were also observed in the peripheral region. The distal end of the TB formed a neck-like region in transition to the MO (Fig. 1, D and I). In the MO, short microtubules (L_{MO_MT} = 221 ± 88 nm, n = 352 microtubules from four cells) were organized into two-line arrays with the electron-dense materials in between (Fig. 1 C), together forming a nanoscopic cytoskeleton. We noted that microtubules in the TB never extended into the MO (zero in four complete sets of volume data), suggesting that microtubules in the outer segment are organized into two arrays.

In leg receptors, we collected mostly lateral sections for technical reasons (see Materials and methods). The outer segment of leg receptors showed an almond-shaped morphology (Fig. 2 C). Similar to haltere receptors, microtubules in the MO were embedded in the electron-dense materials. However, no neck-like region was found. Because lateral sections do not allow direct length measurement (Fig. S2), we measured the spatial distribution of microtubule ends in the outer segment. The distal ends of microtubules accumulated at two regions: the distal tip of the outer segment and the border region between the MO and TB (Fig. 2, E and F). The proximal ends also accumulated at two regions: the border region and the proximal end of the TB (Fig. 2, E and F). Close inspection revealed that only a small number of microtubules spanned across the MO and TB (10 in all 140 MO microtubules [7%] in five cells; Fig. 2 F), suggesting that microtubules in the MO were mostly shorter than 800 nm and were separate from those in the TB. Therefore, microtubules in the outer segments of leg receptors are also organized into two arrays, suggesting a common ultrastructural organization of microtubules in fly campaniform mechanosensory cilia.

Having built the structural model, we wondered about the polarity of microtubules in the TB and MO. We first examined the localization of Patronin, a microtubule minus end stabilizing factor (Goodwin and Vale, 2010; Hendershott and Vale, 2014). To avoid overexpression artifact, we generated a Patronin-RFP knock-in (KI) strain (Fig. S1). In the MO of haltere receptors, Patronin signal was enriched at the proximal end (Fig. 3 A), suggesting that the short microtubules in the MO have a “plus end–distal” polarity. In the TB, Patronin showed a diffusive distribution (Fig. 3 A), consistent with most of microtubules being short and the presence of transversely oriented microtubules. To further clarify the polarity of microtubules in the TB, we studied leg receptors (Fig. S1). Patronin showed two enriched signals in the outer segment of leg receptors (Fig. 3 B). The distal signal of Patronin was 0.6 ± 0.1 µm (n = 13 receptors) away from the distal tip of the outer segment, corresponding to the border region between the MO and TB. The proximal signal of Patronin appeared at the proximal end of the TB, where many microtubule ends are located (Fig. 2 E). Therefore, we conclude that most of microtubules in the TB have a plus end–distal orientation.

We also recorded EB1 signals in the outer segment of leg receptors. EB1 comets in the outer segment showed very short trajectories (Fig. 3 C and Video 2), so we cannot measure the moving directions. Instead, we quantified the distribution of EB1 signal along the proximal-distal axis of the outer segment (Fig. 3 D). EB1 signal at the distal end was 135 ± 12% of that at the proximal end (n = 5 cells; P < 0.001; Fig. 3 E). The region with a stronger EB1 signal covered the MO and the distal part of the TB, supporting the idea that most of microtubules in the TB have a plus end–distal polarity. Furthermore, the observation of an EB1 signal at the distalmost tip (Fig. 3 C) and the enrichment of Patronin at the proximal end of the MO (Fig. 3, A and B) suggest that microtubules in the MO also have a plus end–distal orientation.

In contrast to the outer segment, all EB1 comets in the inner segment (n = 5 cells) moved toward the cell body (Video 2), indicating that these microtubules have a “minus end–distal” orientation. This is consistent with previous studies on the polarity of dendritic microtubules in fly neurons (Stone et al., 2008; Wang et al., 2019). We also noted two dot-shaped signals of Patronin appearing around the proximal end of the TB in both haltere and leg receptors (Fig. 3, A and B). Using live-cell imaging and ET, we found that these two dots were outside the sensory neuron and were associated with a group of microtubules in the thecogen cell that wraps the dendrite of the sensory neuron in campaniform mechanoreceptors (Fig. S1).

We then measured the stability of microtubules in the outer segment using FRAP (Fig. 3 F). Leg receptors were used in this experiment due to the convenience of obtaining lateral views. We found that the tubulin signal in the inner segment fully recovered in 3 h, while that in the outer segment recovered to only ∼25% (Fig. 3 G), suggesting that microtubules in the outer segment are more stable. To further justify this conclusion, two issues were considered. First, does the tubulin signal reflect the mass of microtubules or soluble tubulin dimers? We found that the density of microtubules in the outer segment was 156 ± 29 µm⁻² (n = 5 receptors), leading to an estimation that microtubules contributed to ∼98% of the fluorescence tubulin signal (Fig. S2). Similarly, the microtubule density in the inner segment was 25 ± 11 µm⁻² (n = 3 receptors), suggesting that microtubules contributed to ∼87% of the tubulin signal (Fig. S2). Therefore, we conclude that the intensity of the tubulin signal (Fig. 3 G) reflects the mass of microtubules. Second, could the signal recovery in the outer segment be limited by the transportation of GFP-tagged tubulin dimers into the sensory cilium? As a control, the signal recovery of EB1-tandem dimer (td) GFP (dimeric, ∼116 kD), which has a similar molecular weight to GFP-αtub84B/βtub dimer (∼125 kD), was measured. EB1 signal in the outer segment showed a rapid recovery (τ = 2.2 ± 0.3 min, n = 6 cells; Fig. 3 G), which is an order of magnitude faster than tubulin recovery (τ = 48.1 ± 9.8 min, n = 10 cells; Fig. 3 G), suggesting that the transportation of GFP-tubulin cannot be a limiting step. Therefore, we conclude that the fluorescence recovery reflects microtubule turnover, and that the slower turnover rate in the outer segment suggests greater stability.

Based on the structure and dynamics analyses, we conclude that microtubules in the outer segment are mostly unidirectional and stable. In addition, they are organized into two arrays, i.e., a nanoscopic one in the MO and a relatively bigger one in the TB (Fig. 3 H).

Having characterized the structure and dynamics of the microtubule arrays in the outer segment, we wondered how they are formed. In a previous DNA array analysis (Bechstedt, 2006; Bechstedt et al., 2010), kat-60L1, a katanin family member, was found to be highly expressed in fly halteres. Because microtubules in the outer segment, especially those in the MO, are short, and katanin family members are ATPases that sever long microtubules into short ones (McNally and Roll-Mecak, 2018; McNally and Vale, 1993), we explored the functions of kat-60L1.

Using a gal4-driver strain (kat-60L1-gal4), we found that kat-60L1 expressed in haltere campaniform mechanoreceptors and other peripheral sensory cells, including larval dendritic arborization (da) neurons (Fig. 4, A and B). Based on the annotation in Flybase, kat-60L1 has two transcriptional isoforms. To examine their expression patterns, we generated two isoform-specific gal4-driver lines (kat-60L1-short-gal4 and kat-60L1-long-gal4; Fig. 4 A). Kat-60L1-long expressed in peripheral sensory cells and central nervous systems, while kat-60L1-short only expresses in central nervous systems (Fig. S3). Using RT-PCR, we confirmed that only the long isoform can be detected in haltere RNA samples (Fig. 4 C). We then analyzed the subcellular localization of kat-60L1-long. Using a neuronal driver (elav-gal4) and two more specific neuronal drivers (kat-60L1-long-gal4 and dcx-emap-gal4), we found that kat-60L1-long (RFP-kat-60L1-long) showed two subcellular localizations: (1) an enriched signal at the basal body and (2) a diffusive signal in the outer segment (Fig. 4 D and Fig. S3).

To examine if kat-60L1 contributes to the sensory function of campaniform mechanoreceptors, we first performed flight tests (Bechstedt et al., 2010). Two mutant lines of kat-60L1 were used, one null mutant (BE6; Stewart et al., 2012) and one piggyBAC insertion mutant (c01236; Lee et al., 2009). BE6, c01236, and c01236/BE6 mutants all showed reduced flight ability, while expressing kat-60L1-long using a pan-neuronal driver (elav-gal4) or a more specific type I mechanoreceptor driver (dcx-emap-gal4) rescued the behavioral phenotype to a large extent (Fig. 4 E), suggesting that kat-60L1 contributes to fly mechanosensation. Using extracellular recording (Liang et al., 2013), we found that mechanotransduction still occurred in the haltere receptors of kat-60L1 mutants but had a significantly reduced response (∼45% reduction; Fig. 4, F and G), suggesting that kat-60L1 does not directly participate in mechanotransduction but likely contributes to the mechanosensitivity. Similar phenotypes were also observed in leg receptors by recording the force-invoked calcium response in vivo (Fig. 4, H–J). Based on these results, we conclude that kat-60L1-long contributes to fly mechanosensation, and its absence impairs mechanosensitivity.

How does the loss of kat-60L1 impair mechanosensitivity? We first checked the morphology of the sensory cilium in haltere receptors. In wild-type cells, the cross-sectional view of the TB showed a ring shape with a maximum area of 3.3 ± 0.5 µm² (n = 6 halteres, 48 cells), and the MOs showed a band shape (length, 2.0 ± 0.1 µm, n = 6 halteres, 48 cells; Fig. 5, A and B). In c01236/BE6, the TBs had a smaller cross-sectional area (2.1 ± 0.3 µm²), and the MOs were shorter (1.4 ± 0.2 µm, n = 6 halteres, 48 cells, P < 0.001; Fig. 5, A and B). Furthermore, the signals of NompC-GFP in the MO and GFP-tubulin in the outer segment were greatly reduced in c01236/BE6 (Fig. 5, A, C, and D). These phenotypes were confirmed at ultrastructural level using serial block-face imaging (Fig. 5, E and F; and Video 3). The outer segment in haltere receptors of c01236/BE6 showed structural defects: the MO was shorter and wider; the neck region was less obvious; the TB contained a disrupted microtubule array and surprisingly many vesicles (Fig. 5, E and F). Based on these results, we conclude that in c01236/BE6 mutants, the outer segment is structurally disrupted, and the levels of mechanosensory molecules, e.g., microtubule and NompC, are reduced, which could account for the impaired mechanosensitivity. We noted that the cilium of haltere receptors in c01236/BE6 mutants lost ninefold structural symmetry, and the mother centriole showed a broken structure (Fig. 5 E). Further analyses demonstrate that these defects can be observed in haltere receptors but not in leg receptors (Fig. S4), suggesting that they are not the primary consequences of losing kat-60L1, but likely secondary effects caused by haltere-specific reasons.

To understand the functions of kat-60L1 at molecular level, we first examined if it could sever microtubules. Using a microtubule spin-down assay, we found that purified kat-60L1-long protein promoted microtubule disassembly in the presence of ATP (Fig. S3). In ADP condition, kat-60L1-long could bind to taxol-stabilized microtubules but lost the disassembly activity (Fig. S3). Further analyses using the total internal reflection fluorescence (TIRF)–based imaging assay showed that kat-60L1 could rapidly sever surface-bound taxol-stabilized microtubules in the presence of ATP but not ADP (Fig. S3). Therefore, kat-60L1 has an ATP-dependent microtubule-severing activity.

We then analyzed microtubule arrays in the outer segment in c01236/BE6. ET reconstruction showed two major changes (Fig. 5, G–I; and Fig. S5). First, the total mass of microtubules in the outer segment was significantly reduced (wild type, 246 ± 61 µm, n = 6 cells; c01236/BE6, 120 ± 17 µm, n = 6 cells; P < 0.01; Fig. 5 H). Second, microtubules in the mutant outer segment were longer (Fig. 5 I). Many TB microtubules, including several doublets, extended into the MO (Fig. 5 G and Fig. S5). Therefore, the microtubule array in the MO was no longer separate from that in the TB. Consistent with this observation, Patronin signal at the proximal end of MOs was nearly absent (i.e., fewer minus ends; Fig. 5, J and K). Therefore, kat-60L1 regulates the mass and length of microtubules in the outer segment of haltere receptors.

The disrupted ciliary structure in the haltere receptors in c01236/BE6 raises a possibility that the structural defects in the outer segment may result from the leaky ciliary gate. Therefore, to confirm our findings, we analyzed leg receptors in c01236/BE6, where the ciliary structures are intact (Fig. S4). The phenotypes were similar to those observed in haltere receptors. For example, the outer segment was smaller, and the signals of NompC-GFP in the MO and GFP-tubulin in the TB were reduced (Fig. 6, A–D). ET data showed that while the total mass of microtubules in the outer segment was much smaller in c01236/BE6, there was a significantly larger percentage of microtubules spanning across the TB and MO (Fig. 6, E–G). The absence of the distal Patronin signal further confirmed that microtubules are longer in c01236/BE6 (Fig. 6 H). Therefore, kat-60L1 coordinates the mass and length of microtubules in the outer segment of leg receptors. We also examined if the loss of kat-60L1 leads to the same defects in other external mechanoreceptors (e.g., bristle receptors), and similar phenotypes were observed (Fig. S5), suggesting common functions of kat-60L1 in different ES organs.

We noted a significantly increased tubulin signal at the distal region of the inner segment in c01236/BE6 (Fig. 6, A and D), suggesting local accumulation of microtubules. This is confirmed by two additional observations. First, Patronin-RFP showed a more prominent signal in the inner segment in c01236/BE6 (Fig. 6, H and I). Second, the frequency of EB1 comets in the inner segment was increased by 30% in c01236/BE6 (Fig. 6, J and K; and Video 2), indicating more microtubule polymerization events. In all, these results demonstrate that kat-60L1 plays a destructive role in the inner segment, and in its absence, the homeostatic dynamics of microtubules in the inner segment are altered to a status that favors polymerization.

We showed that the loss of kat-60L1 disrupts the microtubule array in the MO, but it remains unclear whether this is simply due to overall shortage of microtubules in the outer segment or because the nanoscopic cytoskeleton in the MO requires short microtubules. To clarify this issue, we studied the knockdown mutants of kat-60L1. In two kat-60L1ⁱ mutant strains, the tubulin signal of the entire outer segment was lower than wild type, but the signal in the MO was reduced to a further lower level than that in the TB (Fig. 7, A–C). ET data showed a significantly larger percentage of microtubules spanning across the TB and MO in kat-60L1ⁱ (Fig. 7, D and E), suggesting that microtubules become longer. Furthermore, in comparison to c01236/BE6, the TB in kat-60L1ⁱ contained more microtubules (Fig. 7 F), but the microtubule array in the MO still had a similar size (Fig. 7, D and F). This observation suggests that despite the increase in the number, long microtubules could not significantly improve structural integrity of the microtubule array in the MO, demonstrating the necessity of short microtubules in forming the cytoskeleton in the MOs.

The above point was further confirmed by the observations in the tissue-specific knockdown mutants of Patronin. In two Patroninⁱ strains, tubulin signals in the TB and MO were weaker than those in wild type (Fig. 7, A–C), consistent with the function of Patronin in stabilizing noncentrosomal microtubules. Interestingly, we noted that in Patroninⁱ, tubulin signal in the MO was higher than that in the TB (Fig. 7, A–C), showing that microtubules in the MO were less susceptible to the reduction in the level of Patronin and suggesting the presence of additional stabilizing factors in the MO. Furthermore, the stronger tubulin signal in the MO in Patroninⁱ was abolished in the double knockdown mutant (kat-60L1ⁱ; Patroninⁱ; Fig. 7, A–C), confirming that kat-60L1 is required to form the microtubule array in the MO. The remaining tubulin signal in the TB of kat-60L1ⁱ was further reduced in the double mutant (Fig. 7, A–C), suggesting that kat-60L1 and Patronin work together to promote the growth of microtubules in the TB. Consistent with this idea, overexpression of Patronin led to the formation of microtubule bundles and disrupted the morphology of the outer segment (Fig. 7 G), likely due to over-stabilizing microtubules, which would interfere with the effects of other microtubule-regulating factors, including that of kat-60L1.

The structural reconstruction reveals that microtubules in the outer segment are highly organized and stable. They form two structurally separate arrays (Fig. 8). In particular, the array in the MO contains mostly short microtubules, which facilitate the structural optimization of the MO and provide anchoring sites to a large number of mechanosensory molecules (e.g., NompC), thereby serving key sensory functions. Furthermore, the microtubule arrays in the TB and MO are differentially regulated, suggesting dedicated patterning and regulatory mechanisms in the outer segment of fly external mechanoreceptors.

The present study provides two implications in understanding how the microtubule arrays in fly sensory cilia are formed. First, a large quantity of microtubules is required to form dense arrays in the outer segment. Kat-60L1 and Patronin together amplify the mass of microtubules. In addition, the presence of a small number of microtubules in c01236/BE6 implies that there is a kat-60L1–independent mechanism (denoted as X1 in Fig. 8) to assemble microtubules. These microtubules may serve as the initial substrates of kat-60L1 to create more templates. Second, the nanoscopic cytoskeleton in the MO determines the number and spatial arrangement of NompC force-sensitive channels that can be accommodated in the MO. Kat-60L1 coordinates the number and length of microtubules to build the cytoskeleton in the MO, which likely facilitates the structural–mechanical design of the MO, for example, providing adequate anchoring sites for NompC or withstanding resting tensions for optimal mechanosensitivity (Sun et al., 2019a). An ensuing issue is how short microtubules in the MO can be stabilized. The lower susceptibility of the MO microtubules to the reduction in the level of Patronin suggests the presence of additional stabilizing mechanisms (denoted as X2 in Fig. 8).

Our results suggest that kat-60L1 has two functions in fly ES cilia. First, kat-60L1 increases the mass of microtubules in the outer segment. This agrees with the proposed function of microtubule-severing enzymes in amplifying the local mass of microtubules (McNally and Roll-Mecak, 2018; McNally et al., 2006; Ribbeck and Mitchison, 2006; Roll-Mecak and Vale, 2006; Srayko et al., 2006). Second, kat-60L1 is required to generate short microtubules. It has been shown that microtubule-severing enzymes sculpt cellular microtubule networks; for example, the noncentrosomal microtubule arrays in plant cells (Lindeboom et al., 2013; Zhang et al., 2013) and the meiotic spindle (McNally et al., 2006; Srayko et al., 2006). Previous studies have also implicated the potential functions of microtubule-severing enzymes in neurons (Sherwood et al., 2004; Stewart et al., 2012; Trotta et al., 2004). However, due to the limited optical resolution, visualizing dense microtubule networks (e.g., in neurons) where the severing activity is potentially important remains challenging (McNally and Roll-Mecak, 2018). ET could be useful in such cases (Srayko et al., 2006). In the present study, our structural reconstruction provides direct evidence to demonstrate the severing activity of kat-60L1 in fly ciliated sensory neurons, thereby providing new insights into the neuronal and ciliary functions of microtubule-severing enzymes.

A potential caveat to the above conclusion is that the loss of kat-60L1 also changes the homeostatic microtubule dynamics in the inner segment, so it is formally possible that the structural and functional defects in the outer segment might be a secondary effect. However, we do not think this is the case based on three reasons. First, in kat-60L1ⁱ, no significant microtubule accumulation was observed in the inner segment (Fig. 7), but the cytoskeleton in the outer segment was also disrupted, suggesting that microtubules in the outer segment are more susceptible to the level of kat-60L1, and that the defects in the outer segment are a primary consequence of losing kat-60L1. Second, more microtubule polymerizations, suggested by the observation of more frequent EB1 comets, are observed in the inner segment in c01236/BE6 (Fig. 6, J and K), suggesting that the level of free tubulin dimers is fairly normal. Because tubulin is transported or diffuses into cilium as dimers (Craft et al., 2015), we argue that tubulin influx into the cilia, which matters for microtubule dynamics in the outer segment, is not down-regulated in c01236/BE6. Third, no major defects in the morphology of dendritic inner segments were observed in c01236/BE6 (Fig. S4), suggesting that the functional defects can be mostly attributed to the defects in the outer segment.

Previous in vitro studies proposed that the cellular effects of microtubule-severing enzymes depend on the ratio between the amount of severing enzymes and the local microtubule dynamics (Kuo et al., 2019; Vemu et al., 2018). Here, we show that in the outer segment, where microtubules are stable and the signal of kat-60L1 is weak, kat-60-L1 acts as a constructor, while in the inner segment, where microtubules are more dynamic and kat-60L1 shows an enriched signal, kat-60-L1 tends to act as a destructor. These results provide an in vivo example that supports the predictions from the previously proposed model (McNally and Roll-Mecak, 2018).

The function of katanin family members has been implicated in ciliogenesis (Dymek et al., 2004; Hu et al., 2014; Sharma et al., 2007), especially in the assembly of central pair microtubules in motile cilia. Fly mechanosensory cilia contain no central pair microtubules (Keil, 1997) but numerous noncentrosomal microtubules. The disruption of the cytoskeleton in the absence of kat-60L1 in fly mechanosensory cilia resembles the loss of central pair microtubules in the katanin mutants of other species. This implies a common component in the assembly mechanism of noncentrosomal microtubules inside cilia, which, however, remains a hypothesis waiting for future studies.

All flies were maintained on standard medium at 23–25°C. w1118 was used as the wild-type strain. Kat-60L1^BE6, a null mutant strain, was from Nina Tang Sherwood (Duke University, Durham, NC; Stewart et al., 2012); Ana1-tdTom, B9d1-rfp, and Mks1-gfp strains (endogenously tagged) were provided by Tomer Avidor-Reiss (University of Toledo, Toledo, OH; Basiri et al., 2014). An endogenously tagged Shot-gfp strain and uas-rfp-kat-60L1-long (Sun et al., 2019b) were from Jose Pastor (Tsinghua University, Beijing, China). kat-60L1^i-32506 (BDSC 32506), kat-60L1^i-36866 (BDSC 36866), c01236 (BDSC 10430), elav-gal4 (BDSC 8765), uas-gfp-αtub84B (BDSC 7374), uas-EB1-gfp (BDSC 35512), uasp-gfp-cnn1 (BDSC 7255), uas-CD4-tdTom (ch2) (BDSC 35841), uas-CD4-tdgfp (BDSC 35836), and uas-gCamp6s (BDSC 42746) were from the Bloomington Drosophila Stock Center. Patronin^i-108927 was from the Vienna Drosophila Resource Center (VDRC 108927). Patronin^i-1753 was from the Tsinghua Fly Center (THU 1753). kat-60L1-gal4, kat-60L1-long-gal4, and kat-60L1-short-gal4 were generated by cloning the corresponding promoter sequences (Fig. 4 A) into the pptGal4 vector, which were then used to generate the promoter-gal4 strains. To generate the dcx-emap-gal4 strain, a 1.5-kb upstream sequence in front of the first exon of dcx-emap (CG42247) gene was cloned into the pptGal4 vector. To generate the uas-Patronin-GFP strain, the cDNA of Patronin was cloned into the pAWG vector (Drosophila Gateway Collection). To generate the uas-mCherry-αtub84B strain, the cDNA of αtub84B was cloned into the pMT-mCherry vector, and then the mCherry-αtub84B sequence was subcloned into pHisc-RA-RFP. To generate the uas-CD4-tdTom (ch3) strain, the sequence of CD4-tdTom was PCR amplified from strain 35841 (Bloomington Drosophila Stock Center) and cloned into the pUAST-CRT-I-yw2 vector (Xue et al., 2014).

EB1-tdgfp-KI and patronin-rfp-KI were generated using the CRISPR/Cas9-based strategy (Ren et al., 2013). The EB1-tdGFP line was generated by directly knocking in tdGFP in the carboxyl terminus of endogenous EB1 with homology-directed repair recombination. The Patronin-RFP line was generated by directly knocking in RFP in the carboxyl terminus of endogenous Patronin with homology-directed repair recombination. Briefly, tdGFP or RFP was first cloned into the pBluescript vector. We used fly genomic DNA as template to amplify the recombination arms, which were then cloned into the donor plasmid. To increase the efficiency, we designed two single-guide RNAs for each gene and cloned them into the U6B-driven vectors. We then injected the injection mix containing the donor plasmid and two single-guide RNA–containing vectors into the embryos of the nos-cas9 strain. G1 and G2 progenies were crossed with the balancer strains. Genomic DNA from individual flies (both G1 and G2) were extracted and screened using PCR. The correct KI lines were further confirmed by DNA sequencing, and lines with no obvious defects were chosen for further experiments.

NompC-gfp-KI was generated by directly knocking in GFP in the carboxyl terminus of endogenous NompC with homology-directed repair recombination (Fig. S1). The gRNA designed to target the NompC gene (gRNA: 5′-GGT​GGG​CAG​ACG​AGC​CAT​CA-3′) was cloned into the pCFD5 vector as previously described (Port and Bullock, 2016) to get the construct pCFD5-NompC-gRNA. A 1.8-kb sequence upstream of insertion site was ligated with GFP sequences via overlapping PCR. Then, two sequences (2.5 kb and 2.2. kb) corresponding to the 5′ and 3′ sides of the targeting site were inserted flanking the RFP cassette to obtain the final construct pMV-NompC-HR (Xue et al., 2014). To get the GFP KI fly, pCFD5-NompC-gRNA and pMV-NompC-HR were coinjected into the nos-Cas9-attp2 strain as previously described (Ren et al., 2013). Flies with RFP markers were picked as the positive candidates. Finally, six independent lines were confirmed by DNA sequencing.

High-pressure freezing fixation of fly halteres and legs was developed based on a previous protocol (Sun et al., 2019a). The dissected halteres and legs were submerged in 0.1 M phosphate buffer containing 20% BSA (wt/vol; Sigma-Aldrich; A9647) and 0.05% pluronic acid (wt/vol; Sigma-Aldrich; P2443). The dissected tissue was then sucked into a cellulose capillary tube (Leica; 16706869), which was transferred to a 100-µm-deep membrane carrier (Leica; 16707898) and cryoimmobilized using a Leica EMPACT2/RST or Leica EM HPM100 high-pressure freezer.

Freeze substitution was performed in anhydrous acetone (Electron Microscopy Sciences [EMS]; 10015) containing 1% OsO₄ (EMS; 19110), 0.5% glutaraldehyde (EMS; 16220), 0.1% uranyl acetate (Polysciences; 21447), and 4% pure water with a Leica EMAFS2 automatic freeze substitution device. The samples were kept at −90°C for 40 h, warmed up at a rate of 5°C per hour, incubated at −30°C for 8 h, and warmed up at the rate of 5°C per hour to 0°C. After these steps, the samples were rinsed three times in anhydrous acetone at room temperature and infiltrated with Araldite-epon (EMS; 13940): anhydrous acetone mixtures of 1:3, 1:1, and 3:1 for 1 h at each step, followed by 100% Araldite-epon infiltration overnight. Finally, the samples in fresh Araldite-epon were polymerized at 60°C.

The serial sections (250 nm) were prepared using a Leica Ultracut UCT or Leica EM UC7 microtome and collected on Formvar-coated copper slot grids. For haltere receptors, cross-sections were collected to facilitate data after processing, e.g., microtubule tracing and alignment. For leg receptors, we collected consecutive lateral sections in most cases because the exoskeleton of fly legs often separated from the embedding resin, and the cross-sections, usually with a small tissue area, had a large chance of getting lost or distorted. Post-staining was performed with 2% uranyl acetate in 70% methanol, followed by 0.4% lead citrate (EMS; 17900). 15-nm gold nanoparticles (BBI Solutions; EMGC15) were added to both sides of the sections as the fiducial markers. The dual-axis tilt series ranging from −60° to 60° were acquired using a FEI Tecnai F30 or FEI Tecnai F20 electron microscope (Thermo Fisher Scientific; formerly FEI). An FEI Tecnai F30 electron microscope was equipped with an Axial Gatan US1000 CCD camera and controlled with SerialEM automated acquisition software (Mastronarde, 2005). An FEI Tecnai F20 electron microscope was equipped with a Gatan US4000 (895) CCD camera and controlled with FEI Xplore 3D TEM tomography software.

The sample preparation for FIB/scanning electron microscopy was similar to that used for ET except for the embedding medium (Durcupan ACM; Sigma-Aldrich, 44610). For serial FIB milling and scanning electron miscroscopy imaging, a layer of block surface was milled by gallium ion beam, and the block surface was imaged using an electron beam with 2-kV acceleration voltages, 0.4-nA current, and 8-µs dwell time on a FEI Helios NanoLab G3 UC FIB-SEM. After volume data collection, the images were imported into Amira, aligned using the DualBeam 3D Wizard module, and exported as a stack of images in TIF format. The image stacks were then used for structural segmentation.

The tomography data were processed and analyzed as previously described (Sun et al., 2019a). Tomograms were reconstructed with the IMOD software package (v4.7; Mastronarde, 1997). To stitch the adjacent sections in the z axis, microtubules were traced and used as landmarks. These microtubules were first traced using cylinder correlation and trace correlation lines modules in Amira 6.0 (Thermo Fisher Scientific). The alignment and stitching were performed using the SerialSectionStack module in Amira. The structural segmentation and 3D surface generation for both ET and FIB/scanning electron microscopy data were performed in Amira. All structural measurements in 3D space were performed in Amira.

To perform electrophysiology recording of the haltere campaniform receptors, a decapitated adult fly (2–3 d old) was immobilized by wax on a glass slide mounted on a ball-and-socket stage (Newport; M-RN-86). A tungsten electrode was placed near the haltere nerve. The reference electrode was inserted in the thorax. A sharp probe was mounted onto a piezo stack actuator (Physikinstrumente; P216.90), which was driven by a piezo amplifier/controller (Physikinstrumente; E-481 PICA Piezo High-Power Amplifier/Controller,) to apply the mechanical stimuli to the campaniform receptor array in the pedicel of the haltere. We applied a square wave stimulation at 100 Hz. The voltage difference between the reference and recording electrodes was amplified with a differential amplifier (WPI; DAM50 Bio-Amplifier). Labview was used to program the software to control the stimulation and record the signal. The data were sampled at 10 kHz and stored on a computer. Offline data analysis was performed with MATLAB.

An adult fly (2–3 d old) that expressed GCaMP6s (Ca²⁺ indicator) was immobilized on a 35-mm dish, and its legs were kept stretched using a transparent tape. A glass probe with a spherical tip (60 µm in diameter) was mounted on a piezo actuator (SuZhou Micro Automation Technology Co., Ltd.; PZT 150/7/60 VS12) and was used to apply compressive force (20-µm step) onto the outer segment of leg receptors. The force-induced change in the calcium signal in the soma of leg receptors was then recorded using a spinning-disk confocal microscope equipped with a 100×UPlanSApo objective (NA 1.40; Olympus). The amplitude of the calcium signal was quantified to measure the force-induced responses.

Total fly RNA was extracted with the RNeasy Kit (Qiagen). The cDNA was synthesized using SuperScript III Reverse Transcription (Thermo Fisher Scientific). All PCRs were performed with Phusion DNA polymerase (New England Biolabs).

Flight tests were performed as previously described (Bechstedt et al., 2010). Adult flies (2–3 d) were examined for wing defects and then allowed to recover from carbon dioxide overnight. Standard cylinders coated with paraffin oil were used. Fly vials were turned upside down in a funnel through which flies can fly into the cylinder. When they hit the wall of the cylinder, the flies would get stuck on the wall. The position of each fly was recorded and used for statistical analysis.

Porcine tubulin was purified from brain homogenates by cycles of polymerization/depolymerization followed by purification on the tumor overexpressed gene affinity column and with a final subsequent cycle of polymerization/depolymerization (Gell et al., 2011; Widlund et al., 2012). Labeling of tubulin with tetramethylrhodamine (TAMRA) was performed as previously described (Gell et al., 2010, 2011). To prepare taxol-stabilized microtubules, 40 µM tubulin (10% TAMRA labeling) in BRB80 containing 1 mM GTP, 4 mM MgCl₂, and 4% DMSO was first incubated on ice for 5 min and then polymerized at 37°C for 30 min. Then, 400 µl of warm BRB80 buffer containing 20 µM taxol was added to stop the reaction and stabilize the microtubules. The taxol-stabilized microtubules were then collected in BRB80 with 20 µM taxol using an Airfuge (Beckman). The stabilized microtubules were subjected to microscope analysis and microtubule spin-down analysis to examine the function of purified kat-60L1.

The spin-down assay was performed as previously described (Varga et al., 2006). Briefly, taxol-stabilized microtubules were diluted in BRB80 supplemented with 20 µM taxol and 0.001% Tween-20 to 0.5 µM tubulin dimer concentration and incubated with 300 nM kat-60L1 (in the presence of 1 mM ATP or ADP) for 30 min at room temperature. Finally, the samples were pelleted using an Airfuge (Beckman). The supernatant and pellet were then analyzed by SDS-PAGE.

The in vitro microtubule assay was performed as previously described (Gell et al., 2010). Briefly, taxol-stabilized microtubules were immobilized on the surface of a silanized coverslip using an anti-TAMRA antibody (Thermo Fisher Scientific; A6397). 300 nM kat-60L1 was added into the reaction in the presence of 1 mM ATP or ADP. The imaging buffer was BRB80 supplemented with 80 mM D-glucose, 0.4 mg/ml glucose oxidase, 0.2 mg/ml catalase, 0.8 mg/ml casein, 1% β-mercaptoethanol, and 0.001% Tween-20. TIRF imaging was performed at 35°C using an Olympus Cell TIRF system with an Andor 897 Ultra EMCCD and a 100×/1.45 NA TIRF objective.

The coding region of the kat-60L1-long was modified by addition of an MBP tag followed by a GFP tag on the amino terminus. Then, kat-60L1-long was expressed in Spodoptera frugiperda cells using a modified Bac-to-Bac system (Invitrogen). The MBPTrap column (GE Healthcare) was used to purify the recombinant protein according to the manufacturer’s instructions. Briefly, the cells were lysed with the lysis buffer (80 mM Pipes, 1 mM EGTA, 1 mM MgCl₂, 100 mM KOH, 100 mM KCl, 1 mM DTT, 0.1% Tween-20, and 0.1 mM ATP, pH 6.9). The lysate was first clarified by centrifuging at 80,000 rpm for 15 min. The supernatant was filtered through a 0.22-µm syringe filter and then applied to the MBPTrap column. The column was washed with the lysis buffer. Then the column was equilibrated back to the lysis buffer, and the protein was eluted with 10 mM maltose in lysis buffer. Finally, the protein was subjected to buffer exchange to remove maltose (80 mM Pipes, 1 mM EGTA, 1 mM MgCl₂, 100 mM KOH, 100 mM KCl, and 0.1 mM ATP, pH 6.9) and used immediately after the purification (without freezing) to avoid protein aggregation. The purified proteins were analyzed using SDS-PAGE (Coomassie blue staining).

Data plotting and statistical analysis were performed using Igor Pro 6.11 (WaveMetrics). Box plots (Fig. 4 E and Fig. 5 I) were shown with minimum, lower quartile, median (line in the box), mean (circle), upper quartile, and maximum.

The position of individual microtubule ends was determined by inspecting ET volume data using Amira 6.0 (Thermo Fisher Scientific; Fig. S2). For leg receptors, the distalmost tip was set as the reference point. The distance between each microtubule end and the reference point was measured to locate the end. When positioning a microtubule end, we also recorded whether it was the proximal end or the distal end of a microtubule (Fig. 2 E).

To measure the local fluorescent signal of a protein (e.g., GFP-tubulin, NompC-GFP, or Patronin-RFP), the regions of interest were first selected. The shape (e.g., area) and fluorescence intensity of the selected regions were then measured using ImageJ.

To measure the density of microtubules in volume data, the area of regions of interest was first measured, and then the number of microtubules inside this area (e.g., TB) was counted using Amira 6.0 (Thermo Fisher Scientific). The density was calculated as the number of microtubules divided by the area.

The width of the cytoskeleton in MO was measured at 500 nm away from the distalmost tip (i.e., the reference point) in the ET reconstructed tomograms using Amira 6.0 (Thermo Fisher Scientific).

The number of MO microtubules was directly counted in ET volume data (Fig. 2 E, Fig. 6 E, and Fig. 7 D). Based on our analyses on wild-type cells, MO microtubules in leg receptors were rarely longer than 800 nm, so the MO microtubules whose proximal end was farther than 800 nm away from the distalmost tip (i.e., the reference point) were marked as one spanning across TB and MO. The number of such microtubules was divided by the total number of MO microtubules to get the ratio (Fig. 2 F, Fig. 6 G, and Fig. 7 E).

Fig. S1 shows the cartoon schematics of two KI strains and live-cell imaging of fly leg campaniform mechanoreceptors. Fig. S2 shows the representative images that were used to map microtubule ends in the outer segment and to measure microtubule density in the inner segment. Fig. S3 shows the expression pattern, subcellular localization, and severing activity of kat-60L1. Fig. S4 shows the data on the ciliary and dendritic structures in haltere and leg receptors. Fig. S5 shows the ultrastructure of the sensory cilium in the haltere pedicellar campaniform mechanoreceptor in c01236/BE6. Video 1 shows the stitched ET volume data and the segmented microtubule model of the outer segment in a wild-type haltere pedicellar campaniform mechanoreceptor. Video 2 shows EB1 comets in the dendritic inner and outer segments of wild-type and mutant leg campaniform mechanoreceptors. Video 3 shows the segmented models of the dendritic outer segments in wild-type and mutant receptors.

The authors thank Jonathon Howard and Susanne Bechstedt for their initial contribution to this work; Jana Meissner and Tobias Fürstenhaupt for their technical assistance; Nina Tang Sherwood for the Kat-60L1^BE6 strain; and Tomer Avidor-Reiss for Ana1-tdTom, B9d1-RFP, and Mks1-GFP strains. Special thanks to the electron microscopy facility in Tsinghua University.

We acknowledge our funding from the National Key R&D Program of China (2017YFA0503502), National Natural Sciences Foundation of China (31922018 and 32070704), Qingdao National Laboratory for Marine Science and Technology (QNLM2016ORP0301), Tsinghua-Peking Center for Life Sciences, and Beijing Advanced Innovation Center for Structural Biology, Tsinghua University.

The authors declare no competing financial interests

Author contributions: L. Sun, L. Cui, Z. Liu, and X. Liang performed ET and light microcopy experiments. Q. Wang, Z. Xue, M. Wu, T. Sun, D. Mao, J. Ni, and J.C. Pastor-Pareja made the transgenic flies. Z. Liu and X. Liang performed electrophysiology recording and calcium imaging. X. Liang, L. Sun, and L. Cui wrote the paper.